ANSÄTZE ZUR MASCHINELLEN ERKENNUNG DER ALTERSGEMÄSSEN ENTWICKLUNG DES KINDLICHEN EEG IM RAHMEN EINER 5-JAHRES-LÄNGSSCHNITTSTUDIE

Abstract
A study was initiated with a view to early detection of dysfunction of the CNS and to observe the physiological development of the CNS by EEG obtained in the age groups of new-born, half, 1, 2, 3, 4 and 5 yr. This was supported by standardized clinical examinations for neurological and somatic findings and by standardized tests of psychomotor and intellectual development. The data as used for automatic processing consisted of 110 EEG in the age groups of half, 1, 2, 3 and 4 yr. Bipolar EEG were recorded using ten-twenty method with the following leads: F4-C4, P4-O2, Fe-C3 and P3-O1. For feature extraction 3 methods of data reduction were applied, i.e., interval-amplitude, spectral analysis and the autoregressive model. From the features obtained, age-specific frequency parameters were selected by statistical methods; further evaluation was performed by cluster and discriminant analysis. The former showed an unequivocal case grouping for each age group. Using unmatched samples of only clinically healthy children from the 3 age groups of 1/2, 1 and 2 yr, linear discriminant analysis were applied to the parameters of the 3 methods of data reduction. This procedure yields in mean recognition rates of 95%, 98% and 98%, respectively, in hold-one-out classification. Similar results are obtained with samples from 4 or 5 age groups (1/2-4 yr). The results of this type of automatic EEG analysis show that discriminant analysis can be used to allocate EEG to specific age groups; it may readily be ascertained whether differences between chronological age exist and development age as evidenced by the EEG.